Skip to main content

Full text of "BSTJ 59: 8. Oct 1980: High-Speed Digital Lightwave Communication Using LEDs and PIN Photodiodes at 1.3 um. (Gloge, D.; Albanese, A.; Burrus, C.A.; Chinnock, E.L.; Copeland, J.A.; Dentai, A.G.; Lee, T.P.; Li, Tingye; Ogawa, K.)"

See other formats


Copyright © 1980 American Telephone and Telegraph Company 

The Bell System Technical Journal 

Vol. 59, No. 8, October 1980 

Printed in U.S.A. 



High-Speed Digital Lightwave Communication 
Using LEDs and PIN Photodiodes at 1 .3 /im 

By D. GLOGE, A. ALBANESE, C. A. BURRUS, E. L. CHINNOCK, 

J. A. COPELAND, A. G. DENTAI, T. P. LEE, TINGYE LI, 

and K. OGAWA 

(Manuscript received April 9, 1 980) 

At the wavelength of 1.3 /im, fiber loss and dispersion are suffi- 
ciently small that many lightwave communications applications can 
use simple and reliable led transmitters and pin photodiode receiv- 
ers without avalanche gain. Bit rates can be as high as several 
hundred Mb/s. With new high-speed devices based on III-V semi- 
conductors and microwave silicon ic technology, we have designed 
two fully retimed optical regenerators that operate at 1.3 [im and at 
bit rates of 44.7 and 274 Mb/s to study the potential of the led -pin 
approach. A detailed analysis of the baseband characteristics of the 
led, the fiber, and the receiver leads to an overall equalization 
approach that minimizes receiver noise. The success of this perform- 
ance optimization is corroborated by bit- error measurements under 
simulated system conditions. The results suggest repeaterless opera- 
tion over distances up to 24 km at 44. 7 Mb/s and 8 km at 274 Mb/s 
for a cable loss of 1 dB/km and a bandwidth of 1000 MHz -km. The 
use of lasers in such multimode fiber sytems would permit larger 
margin allocations and penalties than those chosen for led systems, 
but would not lead to substantially longer repeater spacings. 

I. INTRODUCTION 

Optical-fiber communication systems that use leds as sources and 
pin photodiodes as detectors offer the attractive features of reliability, 
simplicity, and low cost. Operation of such systems at high digital 
speeds is now made possible by the availability of InGaAsP leds and 
detectors, and of low-loss fibers that exhibit minimal material disper- 
sion in the wavelength region near 1.3 /xm. 1 The bandwidth-distance 
product at the wavelength of minimum dispersion is inversely propor- 
tional to the square of the spectral width of the source and, for 
germania-doped silica fibers, is 3.5 GHz • km or more if surface-emitting 

1365 



leds are used. 2 Thus high-speed digital systems of a few hundred 
Mb/s can be operated over repeater spans of 5 to 10 km. 

An earlier paper 3 presented a discussion of material-dispersion-lim- 
ited operation of fiber systems that use leds and reported an experi- 
ment that showed the effects of material dispersion on a 137-Mb/s 
data link operating at X = 0.89 or 1.20 jum. In the present paper, we 
discuss further the application of leds and pin photodiodes in high- 
speed digital fiber systems and describe two optical repeater experi- 
ments (with bit rates of 44.7 and 274 Mb/s) at X - 1.3 /rai using 
InGaAsP leds as transmitters and InGaAs pin photodiodes followed 
by GaAsFET preamplifiers as receivers. Experimental results are com- 
pared with theory. 

II. SYSTEM CONSIDERATIONS 

For high-speed digital applications, the characteristics of the led 
that are important are its reliability, output power, modulation band- 
width, spectral width, and temperature behavior. The reliability of 
high-radiance, surface-emitting InGaAsP leds is now well established, 
as results from accelerated-aging tests at elevated temperatures project 
a mean life in excess of 5 X 10 9 hours for room-temperature operation. 4 
Sufficient power at A = 1.3 jum is available from these devices: peak 
powers of 200 /xW have been coupled into large-core step-index optical 
fibers. 4,5 The modulation bandwidth of an led depends on the device 
geometry, the current density, and the doping concentration and width 
of the active region; a high doping concentration or a narrow active 
region results in large bandwidth but yields low output power. 6 

The fall-off of the modulated output power P s (f) of an led as a 
function of the modulating frequency /can be described by the function 

H a {f) m (1 + if/b)-\ (1) 

where b is the modulation bandwidth of the led defined at the half- 
power point (3-dB electrical). For high-radiance surface-emitting 
diodes, typical values of the average power P s that can be coupled into 
standard multimode fibers (50-jum core diameter, 0.23 n.a.) are P a = 50 
juW for b = 50 MHz. In general, P a is proportional to b~" where cSl. 
For moderately fast, double-heterostructure GaAlAs surface emitters, 
v = % is a typical value. 6 leds with bandwidths in excess of 1 GHz 
have been fabricated; 7 the output of these very fast diodes tends to 
decrease as 1/6. 

Compared to an injection laser, the output power of the led is 
rather insensitive to temperature variations. Observed data on the 
performance of the led used in our repeater experiments are given in 
Section III. 

Microwave GaAsFETs are ideally suited for use as low-noise pream- 

1366 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980 



plifiers following the pin photodetector in an optical receiver because 
they have low input gate capacitances and high-transconductances. 8 ' 9 
The sensitivity of an optical receiver using state-of-the-art pin photo- 
diodes and GaAsFETs can be made to fall within a few decibels of that 
of the typical silicon avalanche photodiode receiver used presently in 
fiber systems that operate at X = 0.85 /xm. 10,11 At longer wavelengths, 
significant bulk leakage currents at voltages approaching breakdown 
severely limit the use of avalanche gain as a low-noise amplification 
process. For this reason, very low-noise preamplifiers will play a key 
role in future long-wavelength systems. 

We make use of Personick's theory 12 to calculate the receiver sen- 
sitivity, which can be expressed in terms of N r , the average number of 
signal photoelectrons per bit required at the detector to yield an error 
probability of 10~ 9 . N r is given by three terms which account for 
Johnson, shot and fet channel noise: 



qB [ R L g m _ J 



1/2 

(2) 



where 



q = electronic charge 

B = bit rate 

k = Boltzmann's constant 
R L = equivalent input resistance 
II = total leakage current (photodiode and fet) 

r = noise factor associated with channel noise in the fet and is 
taken to be 1.15 for a GaAsFET with negligible gate noise 1213 

T = absolute temperature in degrees Kelvin 
Ct = total capacitance (photodiode, fet and stray) 
g m = transconductance of the fet. 

The coefficients <Ji and J% account for the noise distribution in the 
receiver channel and are discussed subsequently. 

Figure 1 presents receiver sensitivity estimates using eq. (2), assum- 
ing Rl to be infinite and the best available devices to have g m = 30 mS 
and Ct = 1.5 pF. For II = 0, N r is proportional to Ct/ "fgin- The dashed 
diagonal lines help to determine the sensitivity in terms of the average 
optical power 

P r = NrhcB/X-q (3) 

required at the receiver, where h is Planck's constant and c is the 
speed of light. They are computed for a wavelength A = 1.3 /xm and an 
external quantum efficiency tj = 0.7. 

Figure 1 also indicates the performance degradation resulting from 
leakage or dark current in the fet and the photodiode. Such currents 

HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1367 



g m = 30 mS 
7J = 0.7 
X = 1.3 uri 




10 100 

BIT RATE IN MEGABITS PER SECOND 



1000 



Fig. 1 — Average number of photoelectrons per bit vs bit rate according to eq. (2) 
required for error probability of 10~ 9 ; optical receiver uses piN-GaAsFET front-end with 
total input capacitance Ct, leakage current h, and transconductance g m = 30 mS. 
Dashed lines indicate average optical power required at the receiver for quantum 
efficiency tj = 0.7 and wavelength X = 1.3 /xm. 



can be as high as 200 nA at typical elevated operating temperatures 
(see Sections III and IV). Another noise phenomenon, 1//" noise, is not 
indicated in Fig. 1, but is discussed briefly at the end of Section IV. A 
similar performance evaluation on the basis of Personick's theory has 
been conducted for receivers using the most advanced silicon bipolar 
transistors (Ct = 2pF, current gain /? = 300). The results are listed in 
Table I for some bit rates of interest. Also listed is the expected 
performance for the case of an avalanche photodiode followed by a 
GaAs fet amplifier. In this case, we assume a maximum bulk leakage 



1 368 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1 980 



Table I — Comparison of expected sensitivity of 
three types of receivers 

Bit Rate 44.7 Mb/s 274 Mb/s 1096 Mb/s 

PIN/(Si)Bipolar -45 dBm -38 dBm -32 dBm 

PIN/(GaAs)FET -50 dBm -40 dBm -32 dBm 

APD/(GaAs)FET -43 dBm -38 dBm 



current of 200 nA before multiplication and a carrier-ionization ratio 
of 4:1. 

The use of J 2 and J 3 in eq. (2) is based on the assumption that the 
digital information arrives undistorted at the receiver input and is 
transformed by the receiver filter function H r (f) for minimum noise 
and intersymbol interference. In this case, 



Jn+2 'B 



H r 



df. (4) 



Table II lists J n for a non-return-to-zero signal format and a raised- 
cosine signal spectrum at the receiver output, in which case H r = {irf/ 
2B) cot (irf/2B). 

In general, the signal arriving at the receiver has been modified by 
the response of the source, H s (f), and the response of the transmission 
medium, H t {f). To compensate for these effects, the receiver must be 
designed with a response H' r = H r /H s H t . As a first example, consider 
the case of an led with response given by eq. (1) and unlimited fiber 
bandwidth. Let fet channel noise be the dominant noise contribution 
as described by the third term in eq. (2). The coefficient J$ in that 
term now becomes Ja + J a (B/b) 2 because H r in eq. (4) is replaced by 
H'r. Hence the required receiver power becomes P' r = P r [l + (Ja/Jz) 
(B/b) 2 ] l/2 where P r is the value previously computed from eqs. (2) and 
(3). This degradation of receiver sensitivity represents the power 
penalty one pays for equalizing the response of the led. 

The available optical signal margin between transmitter and receiver 
is proportional to P s /Pr, where P s is the average power at the trans- 
mitter as defined earlier. For P s proportional to b~ v , the function 

P s /P' r has a maximum at b = B I - - 1 I (J 4 /J 3 ) 1/2 or b = 0.42B if v 



Table II — Values of J n 



Jl 


0.5628 


J, 


0.0868 


J, 


0.0304 


J, 


0.0143 


J, 


0.0079 


Jl 


0.0048 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1369 



= %. The latter condition results in a degradation of receiver sensitivity 
(or an equalization penalty) of 2.4 dB over the values given in Fig. 1. 
A faster led would reduce this penalty, but would provide less output 
power and a smaller overall signal margin. 

Chromatic and intermodal dispersion in the fiber also influence the 
signal. Neglecting nonlinear, coherence and waveguide effects, we can 
model the fiber transfer characteristic by the product of two filter 
functions H c and Ha, one describing material dispersion and the other, 
intermodal delay differences. For a given transmission distance z, the 
material effect can be determined in advance for typical led systems 
and can be compensated for in the receiver filter. At the wavelength of 
minimum material dispersion (—1.3 jum), the transfer function attrib- 
utable to material dispersion is 2 

H c (f) = (1 + iirfz^/WcX 2 )- 1 ' 2 (5) 

for a Gaussian led spectrum having an rms width a = 0.425 fwhm. 
For typical InGaAsP leds with a doping density of 3 X 10 18 cm -3 , the 
rms width is a — A 2 /40 /nm, where A is in /mi. 14 

Figure 2 shows the increase in the required receiver power as a 
result of material dispersion in fibers plotted against the product of bit 
rate B and transmission distance z. The led bandwidth b is used as a 
parameter. The penalties associated with transfer functions given by 
eqs (1) and (5) add nearly linearly. 

Present fiber fabrication processes are not sufficiently precise to 
provide the desired control of intermodal delays (less than ~0.5 ns/km 
total). Thus, unable to anticipate the resulting baseband characteristics 
precisely, we must also allow for an equalization penalty due to 
intersymbol interference in addition to that caused by first-order 
equalization in the receiver. To estimate the latter effect, we use the 
simple functional dependence 

H d (f) = (\ + iSzf/Qy l (6) 

following the usual notation of a bandwidth-distance product Q defined 
at \H d \ - % (3 dB optical). The additional penalty factor [1 + 3(J 4 / 
J3) (zB/Q) 2 ] 1/z can be calculated in a manner similar to that described 
above for the source. 

III. EXPERIMENTAL LEDs AND PIN PHOTODIODES 

The high-radiance light-emitting-diode structure used in this work 
is shown in Fig. 3. The diode was a small-area, etched-well surface 
emitter, described previously, 15 made from a double-heterostructure 
wafer of InP/Ini-^Ga^AsyPi-^/InP grown by liquid-phase epitaxy. The 
composition of the quaternary active layer (x = 0.36, y = 0.80) was 
chosen to produce diodes with an output wavelength of 1.3 jum ± 0.05 

1 370 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1 980 




1000 2000 

z-B IN KILOMETER x MEGABIT PER SECOND 



3000 



Fig. 2 — Increase in optical signal power required at the receiver as a result of material 
dispersion in germanosilicate multimode fibers plotted vs product of distance z and bit 
rate B; b is the led modulation bandwidth. 



/urn, under room-temperature ambient conditions, at an operating 
current of 100 mA dc. This current corresponds to a current density of 
about 5 kA/cm 2 in the primary emitting area, which was about 50 /im 
in diameter. The active quaternary layer, 1 to 1.5 /on thick, was doped 
p-type with zinc to a relatively high carrier concentration of 2 to 4 X 
10 18 /cm 3 . The 1.3-/im wavelength output into the air without antire- 
flection coatings or lenses was about 1 mW at 100 mA. A typical 
multimode fiber (50-/im core diameter, 0.23 n.a.), butt-coupled without 
index matching, collected 10 /xW of that power. The spectral width 
(fwhm) at 25°C and 100 mA was 1145A. The electrical modulation 
bandwidth was ~170 MHz. 

Temperature characteristics of the diodes were measured in the 
range of 20°C to 100°C, and the output power was found to follow 
approximately the relation 

PT = P exp[-(T-r )/T c ], (7) 

where P is the power output at T a (300°K) and the characteristic 
temperature, T c , is 80° K. These results are plotted in Fig. 4, where it 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1371 



CONTACTS 



w/mr//. 




1.3 -Aim LIGHT OUT 




itALU* 



- * W///////M/Mm%//,M. 



. — InP 



, - __- InGaAsP 



■ InP 
SiN 



Fig. 3 — Schematic cross section of the 1.3-/un led. 

can be seen that the output was reduced by 1.5 dB for a temperature 
rise from room temperature to 55 °C. The thermal impedance was 80 
to 100° C/W. The peak emission wavelength was shifted 380A at a rate 
of ~5A/°C toward longer wavelengths by the same temperature in- 
crease. The spectral width varied approximately as A 2 . The electrical 




1.0 








1- 
















=3 0.8 








— 




_ 


o 
















> 








1- 0.6 






"~~~>~»^^^ — 










_j 








w 








"" 0.4 




InGaAsP LED 


- 


0.2 






- 







1 1 1 


1 



20 40 60 80 

TEMPERATURE IN DEGREES CELSIUS 



100 



Fig. 4 — Temperature dependence of wavelength, spectral width (fwhm) and output 
of the 1.3-pm InGaAsP led. 



1 372 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1 980 



modulation bandwidth, however, remained unchanged from 20 to 
100°C. The practical operating life of these specific diodes is unknown, 
but at this writing, similar unencapsulated diodes have been operating 
at 100 mA dc in a room-temperature laboratory environment for over 
23,000 hours with negligible degradation. 

Photodetectors for the experiment were pin photodiodes made with 
In^Gai-jAs lattice matched to InP (x = 0.53). They are described in 
more detail elsewhere. 16 This ternary composition has a bandgap 
energy of 0.75 eV, corresponding to a long-wavelength cutoff near 1.65 
/mi. The attainable leakage current of p-n junctions in this material 
has proved to be one to two orders of magnitude less (better) than 
that associated with similar junctions in germanium, but otherwise the 
photosensitive properties of the two materials are rather similar. 

The InGaAs was grown by a conventional lpe technique to a 
thickness of 6 to 8 /mi on a (100) n-type InP substrate. The epitaxial 
layer was not intentionally doped and, with a suitable bake-out 17 of 
the boat and starting materials before growth, the residual n-type 
carrier concentration of the ternary layer was 1 to 2 X 10 15 /cm 3 . This 
relatively low impurity concentration permits the fabrication of diodes 
with low capacitance and depletion layer widths of 2 to 3 /xm at low 
bias voltages. 1617 

The photodiodes are illustrated schematically in Fig. 5. The p-n 
junction was prepared by diffusion of zinc to a depth of 2.5 to 3.5 fan 
into the grown InGaAs layer. After contacts were applied, small mesas 
with a diameter of about 90 /un were formed by chemical etching and 
windows about 250 jinn in diameter were opened in the substrate 



CONTACTS 

r~ ___ 



13=^ 



.InGaAs 




EPOXY 



1.3-/xm LIGHT IN 



Fig. 5 — Schematic cross section of the back-illuminated long-wavelength pin photo- 
diode. 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1373 



metallization, also by chemical etching, to permit unobstructed illu- 
mination of the junction through the substrate. This back-illuminated 
configuration takes advantage of the transparency of the InP substrate 
to minimize absorption loss and surface recombination and, at the 
same time, to achieve minimum dark current and minimum parasitic 
capacitance: there is no need to enlarge the junction area to provide 
room for contacts or contacting pads. The diodes were operated 
without an anti-reflection coating, and the external quantum efficiency 
approached the transmission coefficient of 0.7 expected at the interface 
between air and InP (refractive index 3.4). 

The 90-/xm mesas had measured junction capacitances of 1 to 1.5 pF 
at zero bias and 0.3 to 0.5 pF at 10 V reverse bias, where the depletion 
width was about 3 /im and the peak electric field was estimated to be 
6.7 X 10 4 V/cm. The best units had dark currents between 1 and 2 nA 
at 25 °C and 10 V reverse bias; the variation of the dark current with 
temperature was exp(-E g /2kT) up to 100°C (see Fig. 6). The dark 
current of the units used in the present repeater experiment ranged 



io- 



i<r° - 



< 10" 



icr s - 



10^ 





1 1 

InGaAsPHOTODIODE 

NO. 139 
DIAMETER 100/im ^^^ 


1 


- 


140 °C^- 


*"■"* 




„*■"*" 100 a c^*-^^ 






< ^-"" 80 'C^ — -"■""^ 






^^■^^"^ 60 °C^- 






^■"^^ 40 "C^^^^ ^. 
^ ^"^ 25 °C^ — "^"^ 






1 1 


1 



5 10 15 

REVERSE VOLTAGE (V) 



20 



Fig. 6 — Dark-current characteristics of one of the best InGaAs diodes in the temper- 
ature range between 25 and 100°C. 



1374 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980 





RECEIVER 








TRANSMITTER 


<j 


i) 






PHASE- 
LOCKED 
LOOP 




P.N^ 




AMPLIFIER 












! AMPLIFIER 


'/ 






EQUALIZER 




/ 




FILTER 






DECIDER 


~\~ 




tf 












i 




AMPLIFIER 








■> 




AGC 





















Fig. 7 — Block diagram of the optical repeater. 

from 5 to 100 nA at room temperature. The reverse breakdown was 75 
to 100 V at 10-juA bias current. The parasitic capacitance of the 
modified silicone-potted "pill" package used in these experiments was 
0.1 to 0.15 pF. The measured rise times of the better units were less 
than 0.2 ns. 

IV. REGENERATOR DESIGN 

To test the feasibility of the system configuration described earlier, 
we assembled two experimental regenerators designed to operate at 
44.7 and 274 Mb/s. Figure 7 shows a block diagram of the essential 
functions. The transmitter module was the same for both regenerators 
and consisted of a two-stage GaAs fet power amplifier driving the led 
previously described. The amplifier produced pulses of 200-mA peak 
current and 500-ps rise time. With this design, it was possible to 
generate pseudo-random optical pulse patterns at bit rates up to 365 
Mb/s. Figure 8 shows the transmitter output at 274 Mb/s. The average 
power launched into a fiber of 50-/xm core diameter and 0.23 n.a. was 
about 5 jtW. 

The receiver module consisted of the linear amplifier and filter, 
decision, and timing recovery circuits. Except for the timing recovery 
function, the 274-Mb/s design made use of high-speed integrated 
circuits based on the microwave isolated-junction technology. The 
timing recovery circuit used was a standard part of the Western 
Electric regenerator designed for the T4M coaxial transmission sys- 
tem. 18 The linear channel consisted of the optical detector, a GaAs fet 
preamplifier described below, an equalizer, a 60-dB booster amplifier 
with automatic gain control, and a multistage lc filter designed to 
produce the desired noise characteristic. 11 The booster and regenerator 
portions of the 44.7-Mb/s regenerator were standard parts presently 
used in the FT3 lightwave transmission system. 19 

The front-end portions of both regenerators consisted of the InGaAs 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1375 



detector previously described, a low-noise dc-coupled feedback ampli- 
fier as shown in Fig. 9 and an rc equalizing circuit. The three-stage 
amplifier was comprised of a GaAs fet and two complementary 
microwave transistors. 11 The critical circuit parameter was the input 
capacitance Ct, which consisted of the photodiode capacitance (0.5 
pF), the fet gate-source capacitance (0.5 pF), and the parasitic circuit 
capacitance (0.5 pF). With Ct = 1.5 pF and g m = 30 mS, the perform- 
ance resulting from fet channel noise can be found in Fig. 1. 

The extraordinary speed of the microwave transistors used {f c = 4 
GHz), as well as a special computer-aided circuit design process, made 
it possible to construct a feedback amplifier useful at bit rates up to 
several hundred megabits per second. This approach assured wide 
amplifier bandwidth, wide dynamic range, and a single-pole character- 
istic which minimized the subsequent equalization required. Figure 10 
shows the test package with coaxial connections. The open-loop gain 
A and the feedback resistance Rl were chosen as indicated in Table 
III. The objective was to maximize the bandwidth of the feedback 
amplifier, which is roughly proportional to A/R l Ct, without adding 
significant front-end noise. The first term in eq. (2) describes the noise 




Fig. 8 — Eye diagram at the transmitter output (B = 21 A Mb/s). 
1376 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980 




Fig. 9 — GhAsfet preamplifier circuit diagram. 

contribution due to R L ; for the two cases listed in Table III, the noise 
contributions are shown in Fig. 1 by the two crosses. 

The gate leakage current of the GaAs fets was a function of 
temperature and gate bias voltage. For an operating bias V G s of —0.5 
V, the best units had a leakage current of 1 to 2 nA which increased to 
5 to 10 nA at 70° C. 20 As in the case of the detector diode, there was a 
significant spread in performance, but most of the results described 
below were obtained with units not specially selected. At bit rates 
below 100 Mb/s, the 1//" noise component of the GaAs fet can impair 
the sensitivity of the pin-fet receiver. We found this degradation to 
be negligible to bit rates as low as 30 Mb/s in all units employed in our 
receiver studies. 

V. EXPERIMENTAL RESULTS AND CONCLUSIONS 

Two system experiments were performed, one at each bit rate, by 
connecting the transmitter and receiver modules shown in Fig. 7 via 
multimode test fibers. Pseudo-random bit patterns were transmitted 
at the design rate and the system was tested for bit errors. Figure 11 
shows the results. The solid lines are best fits to the data points. 

For comparison with theory, one can compute P r from eqs. (2) and 
(3) using g m = 50 mS, C T = 1.5 pF, R L from Table III and I L = 75 nA 
as measured for the combined leakage current of the fet and the 
photodiode. The external quantum efficiency of the detector in the 
44.7-Mb/s receiver was 0.50, while for the 274-Mb/s receiver tj was 
0.66. With these values, we calculate an expected receiver sensitivity 
of -49.7 dBm at 44.7 Mb/s and -39.8 dBm at 274 Mb/s, on the 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1377 




Fig. 10— Front-end module with coaxial test connections. 

assumption that the led and the fiber have infinite bandwidths. 
Actually, the led bandwidth was 170 MHz, while the fibers used had 
bandwidths (3-dB optical) of 45 MHz in the 44.7-Mb/s experiment and 
200 MHz in the 274-Mb/s experiment. As explained in Section III, 
these conditions lead to a modification of the J- values in eq. (2). If one 
uses eqs. (1), (5), and (6) to compute the penalty factors, the expected 
sensitivities become -48.2 dBm at 44.7 Mb/s and -36.8 dBm at 274 
Mb/s. 

The discrepancies from the measured data shown in Fig. 11 are 1.5 
and 0.8 dB. We ascribe the difference at 44.7 Mb/s to induced gate 
noise and to the uncertainty in determining the actual amount of 
leakage current, which in our experiments fluctuated with humidity 
and near-term operating history. The discrepancy at 274 Mb/s is 
probably a consequence of intersymbol interference which was ne- 
glected in our computation. However, the relatively good agreement 



1378 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980 



10" 



i= 10- 



10" 



io- 



,_^_--46.7dBm 



J I I I I L 



J L 



-50 



-45 -40 

AVERAGE OPTICAL POWER IN dBm 



-35 



Fig. 11 — Error rate vs average optical input power measured at 44.7 and 274 Mb/s. 

gives us confidence that the equalization in the receiver can compen- 
sate for the known bandwidth limitations of the optical channel. 

The remaining question of interest concerns the expected repeater 
spacing possible with led-pin systems operating at speeds above 40 
Mb/s. Based on current research, the answer will naturally be con- 
servative, but nevertheless it will provide an interesting benchmark. 
Allowing for component variations and elevated temperatures, we 
assume g m = 30 mS, Ct = 1.5 pF, and h = 200 nA, and compute the 
average optical power P r from eqs. (2) and (3), taking into account the 
led modulation characteristics and fiber dispersion due to chromatic 
and intermodal delays; e.g., we use the appropriate J-coefficients in 
accordance with the actual transfer conditions as stipulated by eqs. 

Table III — Pre-amplifier characteristics 



Open Feedback Band- 
Bit Rate Loop Resistance width 
Mb/s Gain kSl (MHz) 



44.7 
274 



30 

12 



500 

30 



3 
50 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1379 



(1), (5), and (6). This computation converts bandwidth limitations into 
an equalization penalty at the receiver. Lacking more specific infor- 
mation on intersymbol interference, we allow for an equalization 
penalty in decibels equal to the ratio zB/Q, where Q is the bandwidth- 
distance product resulting from intermodal delays. (This assumption 
is based on limited experimental experience and deserves further 
study.) In addition to these receiver sensitivity penalties, there are 
fixed margin (power-loss) allocations at the transmitter and the re- 
ceiver as listed in Table IV. 

The average led output into a typical fiber (50-/im core, 0.23 n.a.) 
is assumed to be —13 dBm between and 50 MHz, and to decrease by 
2 dB between 50 and 100 MHz and by 3 dB/octave thereafter. The 
led bandwidth is chosen for maximum signal margin P a /P r . This ratio, 
reduced by penalties and allocations, determines the transmission 
distance possible with a cable having a given over-all loss. Since P s /P r 
is a function of z because of eqs. (5) and (6), we must seek an iterative 
solution to determine z. 

Some representative results are shown in Fig. 12. Cable loss and the 
bandwidth- distance product associated with intermodal dispersion are 
the two parameters used. We make the conservative assumption that 
the bandwidth decreases linearly with length. The transmission dis- 
tances achievable range from 24 km at 44.7 Mb/s to 8 km at 274 Mb/ 
s for cable having a loss of 1 dB/km and a bandwidth-distance product 
in the 1000-MHz • km range. The good agreement with experimental 
results seems to justify the underlying design rules. The influence of 
induced gate noise and the determination of residual intersymbol 
interference need further study. 

Three interesting conclusions can be drawn from the calculation 
that resulted in Fig. 12: 

(i) Loss limits the achievable repeater spacing more critically than 
bandwidth. For a given application, the bandwidth specification and 
the associated equalization penalty allocation are design variables to 
be determined ultimately by economic considerations. 

(ii) The "loss-limited" repeater spacing of led-pin systems varies 



Table IV — Fixed margin allocations 



Transmitter: 




30 °C temperature rise 


1.5 dB 


Output decrease to end of life 


1.5 dB 


Connectors 


1 dB 


Receiver 




Feedback resistance 


1 dB 


Connectors 


1 dB 


Margin 


3 dB 


Total 


9 dB 



1380 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980 



30 



20 



< 

uj 

°r. 10 



III I 

N. N. 1dB/km 






— ^v ^v 






^\2dB/km y\ 






^ S ^^>». 2000 M Hz- km r'' N^ N. 












1000MHz-km<^ 






44.7 Mb/s 


274 Mb/s 




Vlli I 


V 





40 



60 



80 100 200 

BIT RATE IN MEGABITS PER SECOND 



400 



Fig. 12 — Expected repeater distance versus bit rate for led-pin repeater at 1.3 /un; 
installed cable loss and bandwidth -distance product of fiber are the parameters. 



approximately as the inverse of the bit rate in the range between 100 
and 300 Mb/s. Therefore, since this implies a constant bit-rate-distance 
product, the bandwidth requirements are nearly the same for all these 
bit rates. 

(Hi) For a fiber bandwidth between 1000 and 2000 MHz -km and a 
spliced cable loss of — 1 dB/km and allowing a receiver equalization 
penalty of 2 to 4 dB, it is possible to build high-speed (>100 Mb/s) 
led-pin systems with repeater spacings given in Fig. 12. Since the 
equalization penalty increases very rapidly with repeater spacing, an 
increase in source power (e.g., from the use of a laser) would permit 
larger margin allocations and penalties but would not lead to substan- 
tially longer repeater spacings in high-speed multimode systems. 

VI. ACKNOWLEDGMENTS 

We thank W. R. Northover, M. Saifi, and J. W. Shiever for the test 
fibers used, W. O. W. Schlosser and S. H. Wemple for their help with 
the GaAs fets, W. Kruppa for providing the microwave transistors 



HIGH-SPEED DIGITAL LIGHTWAVE COMMUNICATION 1381 



and integrated circuits, and S. E. Miller for guidance and encourage- 
ment through all phases of this work. 

REFERENCES 

1. Tingye Li, "Optical Fiber Communication— The State of the Art," IEEE Trans. 

Commun, COM-26 (1978), pp. 946-955. 

2. D. Gloge, K. Ogawa, and L. G. Cohen, "Baseband Characteristics of Long- Wave- 

length led Systems," Electr. Lett., 16 (1980), pp. 366-367. 

3. W. M. Muska, Tingye Li, T. P. Lee, and A. G. Dentai, "Material-Dispersion-Limited 

Operation of High-Bit-Rate Optical-Fiber Data Links Using leds," Electr. Lett., 
13 (1977), pp. 605-607. 

4. S. Yamakoshi, M. Abe, S. Komiya, and Y. Toyama, "Degradation of High Radiance 

InGaAsP/InP leds at 1.2-1.3 fan Wavelength," Tech. Digest, Int. Elect. Dev. 
Meeting, Washington, D.C., December 3-5, 1979, pp. 122-124. 

5. R. C. Goodfellow, A. C. Carter, I. Griffith, and R. R. Bradley, "GalnAsP/InP Fast 

High-Radiance, 1.05-1.3 /im Wavelength leds with Efficient Lens Coupling to 
Small Numerical Aperture Silica Optical Fibers," IEEE Trans. Elect. Dev., Ed-26 
(1979), pp. 1215-1220. 

6. T. P. Lee and A. G. Dentai, "Power and Modulation Bandwidth of GaAs-AlGaAs 

High-Radiance leds for Optical Communication Systems," IEEE J. Quantum 
Electr., QE-14 (1978), pp. 150-159. 

7. H. Grothe, W. Proebster and W. Hart, "Mg-Doped InGaAsP/InP l.e.d.s for High- 

Bit-Rate Optical Communication Systems," Electr. Lett., 15, (1979), pp. 702-703. 

8. J. E. Goell, "Input Amplifiers for Optical pcm Receivers," B.S.T.J., 53 (1974), pp. 

1771-1793. 

9. D. R. Smith, R. C. Hooper, and I. Garrett, "Receivers for Optical Communications: 

A Comparison of Avalanche Photodiodes with pin-fet Hybrids," Opt. Quant. 
Elect., 10 (1978), pp. 293-300. 

10. D. R. Smith, R. C. Hooper, R. P. Webb, and M. P. Saunders, "pin Photodiode, 

Hybrid Optical Receivers," Conf. Proc, Optical Communication Conference, 
Amsterdam, September 17-19, 1979, Paper 13.4. 

11. K. Ogawa and E. L. Chinnock, "GaAsFET Transimpedance Front-end Design for a 

Wideband Optical Receiver," Electr. Lett., 15 (1979), pp. 650-653. 

12. S. D. Personick, "Receiver Design for Digital Fiber Optic Communication Systems 

I and II," B.S.T.J., 52 (1973), pp. 843-886. See also R. G. Smith and S. D. 
Personick, "Receiver Design for Optical Fiber Communication Systems," Chapter 
IV in Semiconductor Devices for Optical Communications, H. Kressel, ed., New 
York: Springer Verlag, 1980. 

13. W. Baechtold, "Noise Behavior of GaAs Field-Effect Transistors with Short Gate 

Lengths," IEEE Trans. Elect. Dev., ED19 (1972), pp. 674-680. 

14. T. Takagi, "Spectral Half- Width of Spontaneous Emission of InGaAsP Lattice- 

Matched to InP," Jpn. J. Appl. Phys., 18 (1979), pp. 2017-2018. 

15. A. G. Dentai, T. P. Lee, C. A. Burrus, "Small-Area High Radiance c.w. InGaAsP 

leds Emitting at 1.2 to 1.3 fan," Electr. Lett., 13 (1977), pp. 484-485. 

16. T. P. Lee, C. A. Burrus, A. G. Dentai, and K. Ogawa, "Small Area InGaAs/InP pin 

Photodiodes: Fabrication, Characteristics and Performance of Devices in 274 
Mb/s and 45 Mb/s Lightwave Receivers at 1.3 nm Wavelength," Electr. Lett., 16 
(1980), pp. 155-156. 

17. R. F. Leheny, R. E. Nahory, and M. A. Pollack, "Ino.53Gao.47 As pin Photodiodes for 

Long-Wavelength Fiber Optic Systems," Electr. Lett., 15 (1979), pp. 713-715. 

18. F. D. Waldhauer, "A 2-Level, 274 Mb/s Regenerative Repeater for T4M," Confer- 

ence Record, International Conference on Communications, June 16-18, 1975, 
San Francisco; pp. 48-13-48-17. 

19. T. L. Maione, D. D. Sell, and D. H. Wolaver, "Practical 45 Mb/s Regenerator for 

Lightwave Transmission," B.S.T.J., 57 (1978), pp. 1837-1856. 

20. S. H. Wemple, private communication. 



1382 THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980